Paramutation in Drosophila linked to emergence of a piRNA-producing locus

A paramutation is an epigenetic interaction between two alleles of a locus, through which one allele induces a heritable modification in the other allele without modifying the DNA sequence. The paramutated allele itself becomes paramutagenic, that is, capable of epigenetically converting a new paramutable allele. Here we describe a case of paramutation in animals showing long-term transmission over generations. We previously characterized a homology-dependent silencing mechanism referred to as the trans-silencing effect (TSE), involved in P-transposable-element repression in the germ line. We now show that clusters of P-element-derived transgenes that induce strong TSE can convert other homologous transgene clusters incapable of TSE into strong silencers, which transmit the acquired silencing capacity through 50 generations. The paramutation occurs without any need for chromosome pairing between the paramutagenic and the paramutated loci, and is mediated by maternal inheritance of cytoplasm carrying Piwi-interacting RNAs (piRNAs) homologous to the transgenes. The repression capacity of the paramutated locus is abolished by a loss-of-function mutation of the aubergine gene involved in piRNA biogenesis, but not by a loss-of-function mutation of the Dicer-2 gene involved in siRNA production. The paramutated cluster, previously producing barely detectable levels of piRNAs, is converted into a stable, strong piRNA-producing locus by the paramutation and becomes fully paramutagenic itself. Our work provides a genetic model for the emergence of piRNA loci, as well as for RNA-mediated trans-generational repression of transposable elements.

A paramutation is an epigenetic interaction between two alleles of a locus, through which one allele induces a heritable modification in the other allele without modifying the DNA sequence 1,2 . The paramutated allele itself becomes paramutagenic, that is, capable of epigenetically converting a new paramutable allele. Here we describe a case of paramutation in animals showing long-term transmission over generations. We previously characterized a homology-dependent silencing mechanism referred to as the trans-silencing effect (TSE), involved in P-transposable-element repression in the germ line [3][4][5] . We now show that clusters of P-element-derived transgenes that induce strong TSE 6,7 can convert other homologous transgene clusters incapable of TSE into strong silencers, which transmit the acquired silencing capacity through 50 generations. The paramutation occurs without any need for chromosome pairing between the paramutagenic and the paramutated loci, and is mediated by maternal inheritance of cytoplasm carrying Piwi-interacting RNAs (piRNAs) homologous to the transgenes. The repression capacity of the paramutated locus is abolished by a loss-of-function mutation of the aubergine gene involved in piRNA biogenesis, but not by a loss-of-function mutation of the Dicer-2 gene involved in siRNA production. The paramutated cluster, previously producing barely detectable levels of piRNAs, is converted into a stable, strong piRNA-producing locus by the paramutation and becomes fully paramutagenic itself. Our work provides a genetic model for the emergence of piRNA loci, as well as for RNA-mediated trans-generational repression of transposable elements.
Paramutations have been well described in plants 1,2,[8][9][10][11][12] . The best characterized is the b1 paramutation in maize, which involves a small RNA silencing pathway [13][14][15] , changes in DNA methylation levels and chromatin modifications 16 , and shows full penetrance and stability across generations. Paramutation-like phenomena involving microRNAs have been described in mice 17,18 . However, long-term inheritance of a paramutation through generations has not been reported so far in animals.
In Drosophila melanogaster, transposition of P elements causes hybrid dysgenesis, a syndrome of genetic abnormalities including a high mutation rate, chromosome rearrangements and sterility 19,20 . In natural populations, telomeric P elements inserted in heterochromatic telomere-associated sequences (TAS) are master sites for establishing P-element repression in the germ line [21][22][23] . In laboratory lines (for example, P-1152), P-lacZ transgenes inserted in TAS mimics telomeric P elements by repressing germline expression of reporter transgenes inserted at distant euchromatic sites, through a homology-dependent silencing mechanism, TSE [3][4][5]24 . TSE is strongly sensitive to mutations affecting the piRNA pathway 5,25 . Its establishment involves both genetic and epigenetic components: a chromosomal copy of the telomeric silencer transgene must be either paternally or maternally inherited, and a cytoplasmic component containing small RNAs homologous to the transgene must be maternally inherited 4,5 . In addition to telomeric loci, we found that T-1, a tandem repeat cluster of P-lacZ transgenes inserted in the middle of chromosome arm 2R (50C), can also trigger a strong TSE 7 . T-1 and other P-lacZ clusters inserted at the same locus ( Supplementary Fig. 1) induce ectopic heterochromatin and show variegation of the white gene marker in the eye, a phenomenon termed repeat-induced gene silencing 6,26 . However, T-1 triggers strong silencing of various TSE reporter transgenes in the germ line 7 , whereas the other transgene clusters at this locus, including BX2, which contains the same number of transgene repeats as T-1, did not induce detectable TSE (Supplementary Table 1).
The epigenetic properties of T-1 were analysed together with those of the P-1152 telomeric silencer and the BX2 cluster as controls. T-1 and P-1152 showed typical maternal transmission of TSE: strong repression occurred in the germ line of progeny when the silencer was maternally inherited (Fig. 1a), whereas weak or null repression was detected when the silencer was paternally inherited (Fig. 1b). BX2 showed no repression capacity in these crosses. To analyse the relationship between TSE and piRNAs, we sequenced 19-29-nucleotide RNAs from ovaries of T-1, P-1152 or BX2 females (Supplementary Table 2). Abundant small RNAs matched the T-1 sequences in the library from hemizygous females having inherited the T-1 locus maternally (Fig. 1c), but not paternally (Fig. 1d). Among these species, the 23-28-nucleotide RNAs showed the typical 'ping-pong' signature of piRNA biogenesis 27 , including a bias for a 59 U (1U) and a strong tendency to form sense-antisense pairs with complementarity over their first ten nucleotides ( Supplementary Fig. 2). In addition to piRNAs, short interfering RNAs (siRNAs) have been shown to be produced by previously characterized piRNA loci 28 . Similarly, T-1 produced a significant fraction of 21-nucleotide RNAs (Fig. 1c) that do not show the ping-pong signature of piRNAs and probably correspond to siRNAs ( Supplementary Fig. 3a). In agreement with a previous report 29 , small RNAs with similar features were produced by P-1152 in hemizygous females having inherited the P-1152 locus maternally (Fig. 1f). Homozygous P-1152 females produced about twice as many piRNAs as these hemizygous females ( Supplementary  Fig. 4). Finally, only a very low level of small RNAs was produced that matched BX2 in hemizygous females from the BX2 line (Fig. 1e). Hence, maternal inheritance of T-1, as well as P-1152, is associated with both the production of piRNAs derived from these loci and the capacity of these loci to mediate TSE, thereby linking silencing and piRNAs in this system. We next tested epigenetic interactions between the P-1152 telomeric silencer and T-1, and found that chromosomal and maternally transmitted components of T-1 and P-1152 can complement each other to induce TSE ( Supplementary Fig. 5), consistent with the presence of piRNAs matching P-lacZ sequences in ovaries of both T-1 and P-1152 females.
To investigate possible transfer of epigenetic information between T-1 and the inactive BX2 locus, we crossed hemizygous T-1 females with hemizygous BX2 males, and recovered female progeny that had not inherited the T-1 locus and carrying a paternally inherited BX2 locus (Fig. 2). These females showed marked silencing of the TSE reporter transgene, indicating that the cytoplasm of T-1 oocytes can confer new silencing capacities to the inactive allele of the BX2 locus. This de novo silencing allele will be hereafter referred to as BX2* to differentiate it from the initial BX2 allele never having been exposed to a T-1 cytoplasm.
A BX2* line was established and analysed in successive generations (Fig. 3a). Notably, second generation (G 2 ) BX2* females from test crosses with males carrying a TSE reporter transgene still showed a complete TSE (Fig. 3b). This capacity to mediate TSE was fully maintained over 25 generations of the BX2* line (TSE 5 100%, n 5 4,600). TSE remained very strong between G 32 and G 55 (99.4%, n 5 22,700) showing a reversion rate less than 0.5% per generation at 25 uC (Supplementary Discussion). We conclude that maternally inherited factors from the T-1 strain stably paramutated the BX2 locus.
In contrast to BX2 females, ovaries of G 2 BX2* females contained abundant small RNAs matching the BX2 sequence ( Fig. 3c and Supplementary Table 2) with a profile similar to the one observed in T-1 females (see Fig. 1c). The size distribution of these small RNAs showed a large peak corresponding to 23-28-nucleotide small RNAs with the piRNA ping-pong signature ( Supplementary Fig. 2), as well as a discrete peak corresponding to a 21-nucleotide siRNA-like species of RNAs. Therefore, the acquired capacity of the BX2* allele to mediate TSE correlates with the de novo production of lacZ-derived small RNAs from this locus. Finally, BX2*-derived small RNAs were continuously produced in ovaries over at least 42 generations of a BX2* line ( Fig. 3d and Supplementary Figs 2 and 3). Together, these data indicate that the BX2* paramutation is associated with stable production of high levels of small RNAs from the BX2 locus in ovaries.
We next tested whether the paramutated BX2* allele is paramutagenic. We crossed hemizygous BX2* females with hemizygous naive BX2 males and recovered female progeny having inherited the   . BX2* female progeny having inherited cytoplasm from T-1 mothers (orange background) and a BX2 chromosome from fathers were stained for lacZ. b, Females carrying only the TSE reporter P-1039 were crossed to BX2 males. Female progeny from this cross were stained for lacZ. c, P-1039/ BX2* female progeny from the cross in a showed complete TSE, which was scored as indicated in Fig. 1. P-1039/BX2 female progeny from the cross in b did not show TSE. Controls correspond to crosses between Canton y (devoid of any transgene) or T-1 females with P-1039 males, which resulted in progeny showing null and complete TSE, respectively. Original magnification, 320.

LETTER RESEARCH
cytoplasm of BX2* mothers and the BX2 locus from fathers (Fig. 4a). This BX2 allele was then assessed in generation G 2 for its capacity to silence a TSE reporter transgene in the germline. Notably, we observed a complete TSE (Fig. 4a), indicating that the paternally inherited BX2 allele was paramutated through maternal inheritance of BX2* cytoplasm. This newly paramutated BX2 allele, which corresponds to a second-order paramutation, will be hereafter referred to as BX2* 2 . A BX2* 2 line was established and showed stable TSE over 36 generations (Fig. 4a). Moreover, this line retained the capacity to produce large amounts of BX2* 2 -derived small RNAs after 36 generations (Fig. 4b).
Following an identical mating scheme, BX2* 2 females were able to paramutate a paternally inherited BX2 locus, generating a thirdorder BX2* 3 paramutated allele that showed full TSE capacity over 10 generations. Applying this procedure recurrently, we generated a fifth-order paramutated BX2* 5 allele that showed full TSE capacity ( Supplementary Fig. 6). In conclusion, the conversion of BX2 to BX2* by T-1 maternal cytoplasm has all the properties of a paramutation, because it is stable over generations and the paramutated allele shows secondary paramutagenicity. Interestingly, T-1 also fully paramutated C2, another seven-copy transgene inserted at the same location ( Supplementary Fig. 1), whereas lower-copy-number transgenes at this location were paramutated only transiently (Supplementary Table 3). A similar unstable paramutation interaction was also observed between the non-allelic P-1152 and BX2 loci ( Supplementary Fig. 7).
As paramutation in this system is correlated with the production of BX2*-derived piRNAs and siRNAs, we investigated the effect of aubergine and Dicer-2 loss of function on a paramutated BX2 cluster.  Figure 3 | BX2* paramutation occurs and is associated to the production of small RNAs by the BX2 cluster. a, BX2* lines were established as indicated. Bal1 and Bal2 are balancer chromosomes carrying distinct phenotypic markers. BX2* siblings were crossed at each generation to perpetuate the BX2* line. In addition, BX2* females were crossed at various generations (G n ) to males carrying the P-1039 reporter, to score the TSE of BX2* in the G n11 female progeny. b, TSE in BX2* females from generations G 2 and G 25 , and in progeny of crosses from Canton y , T-1 and BX2 females with P-1039 males as controls. T-1 BX2 G25 indicates that BX2 females inherited cytoplasm from T-1 females 25 generations before the present cross. TSE was scored as indicated in Fig. 1 Figure 4 | Paramutated BX2* is paramutagenic. a, BX2* females were crossed with BX2 males and a BX2* 2 line (second-order paramutation) was established as indicated. Bal1 and Bal2 are balancer chromosomes. BX2* 2 siblings were crossed at various generations to perpetuate the BX2* 2 line. In addition, BX2* 2 females were crossed at each generation (G n ) with males carrying the P-1039 reporter transgene to score the TSE of BX2* 2 in the G n11 female progeny. b, Abundance (graph on the left) and length distribution (black histogram in the middle) of 19-30-nucleotide small RNAs matching the P{lacW} transgene in ovaries from hemizygous BX2* 2 females from generation G 36 . Length distribution of the subset of small RNAs only matching lacZ is shown in the blue histogram on the right.

RESEARCH LETTER
The silencing capacity of the BX2* 2 cluster was completely abolished in homozygous aubergine mutants, whereas strong silencing still took place in Dicer-2 homozygous mutants ( Supplementary Fig. 8). Moreover, the BX2* 2 locus still showed full repression capacity after four generations in a Dicer-2 homozygous mutant context. Hence, the BX2* silencing activity requires piRNAs, whereas neither BX2* activity nor inheritance rely on siRNAs. In maize, paramutation can be induced by a non-allelic transgene producing b1-repeat doublestranded RNA (dsRNA) and siRNAs 15 and epigenetic inheritance of the Kit tm1Alf mutant allele in mice seems to result from paternal as well as maternal transmission of small RNAs 17 . These data indicate that paramutations may in some instances involve small RNAs without interactions between alleles at the DNA or chromatin levels. Our findings that, in Drosophila, the BX2 paramutation is triggered by cytoplasmic inheritance strongly support this view. Finally, we investigated the effect of the paramutation on transcription of the BX2 locus by quantitative polymerase chain reaction with reverse transcription (RT-qPCR). BX2 and BX2* showed similar steady-state levels of both sense and antisense transcripts (Supplementary Fig. 9). This observation suggests that paramutation, rather than increasing the pool of piRNA precursor transcripts, activates their downstream processing into piRNAs. Thus, the maternally transmitted piRNAs could trigger production of primary piRNAs and/or ping-pong amplification of secondary piRNAs in the nuage. As paramutation is accompanied by de novo production of high levels of piRNA, it provides an invaluable model to determine the molecular events involved in the genesis of piRNA loci.

METHODS SUMMARY
All crosses were performed at 25 uC. lacZ expression assays were carried out using X-gal overnight staining 30 . The P-lacZ-white construct (named P{lacW}) contains the P-lacZ translational fusion and is marked by the mini-white gene (Supplementary Fig. 1 and Supplementary Table 4). Small RNA libraries from hand-dissected ovaries were prepared using the Illumina kit and sequenced using an Illumina Genome Analyzer II or an Illumina HiSeq-2000, following the manufacturer's instructions. For library comparisons, read counts were normalized to the total number of small RNAs that matched the D. melanogaster genome and did not correspond to abundant cellular RNAs (ribosomal RNA, transfer RNA and small nucleolar RNAs). Overlap signatures were computed for each sequence data set by collecting the appropriate RNA reads matching P transgenes and calculating overlap frequencies with RNA reads on the opposite strand.
Full Methods and any associated references are available in the online version of the paper.

METHODS
Experimental conditions. All crosses were performed at 25 uC and involved 3-5 couples in most cases. lacZ expression assays were carried out using X-gal overnight staining as described previously 30 , except that ovaries were fixed for 6 min. Transgenes and strains. P-lacZ fusion enhancer trap transgenes P-1152, BQ16, BC69 and P-1039 all contain an in-frame translational fusion of the Escherichia coli lacZ gene to the second exon of the P transposase gene and a rosy transformation marker 31 . The P-1152 insertion (Supplementary Table 4) was mapped to the telomere of the X chromosome (cytological site 1A) and consists of two P-lacZ insertions in the same TAS unit and in the same orientation 5 . P-1152 is homozygous, viable and fertile. BQ16 is located at 64C in euchromatin of the third chromosome 4 (Supplementary Table 4) and is homozygous, viable and fertile. BC69 is inserted in chromosome 2 (Supplementary Table 4) in the first exon of the vasa gene and results in a vasa loss-of-function allele; consequently, it is homozygous, female and sterile. P-1039 is located at 60B on the second chromosome (Supplementary Table 4) and is homozygous lethal. P-1152 shows no lacZ expression in the ovary, BQ16 and BC69 are strongly expressed in the nurse cells and in the oocyte and P-1039 shows strong lacZ staining in numerous tissues including the follicle cells, the nurse cells and the oocyte. P-lacZ clusters. Lines with different numbers of P-lacZ-white transgenes 32 located at cytological site 50C on the second chromosome 6,26 were used (Supplementary Table 4). The transgene(s) insertion site is located near the mRpL53 gene, in an Ago1 intron. This site is not a piRNA-producing locus, as observed for instance in the deep-sequencing data set from P-1152 ovaries (data not shown). The P-lacZwhite construct contains the P-lacZ translational fusion and is marked by the miniwhite gene (P{lacW}, FBtp0000204). BX2 carries seven P-lacZ copies including at least one defective copy inserted in direct orientations. T-1 derives from BX2 following X-ray treatments ( Supplementary Fig. 1). T-1 has chromosomal rearrangements including translocations between the second and the third chromosomes. After overnight staining, weak lacZ expression is detected in the follicle cells of BX2 and T-1 female ovaries, presumably because of a position effect at 50C, but no staining is observed in the germ line (data not shown).
Lines carrying transgenes have M genetic backgrounds (devoid of P transposable elements), as do the multi-marked balancer stocks used in genetic experiments. The Canton y and w 1118 lines were used as controls completely devoid of any P element or transgene. Crosses involving P-1152 were performed with females carrying the telomeric transgenes in the homozygous state (except where indicated), whereas crosses performed with BX2 or T-1 were performed with females carrying the cluster in the heterozygous state (referred to as hemizygous in case of insertions) because of the sterility (BX2) and lethality (T-1) induced by transgene clusters.
Two strong hypomorphic mutant alleles of aubergine induced by EMS were used. Both of them are homozygous, female and sterile, and TSE was previously shown to be abolished by a heteroallelic combination of these alleles 5 . aub QC42 comes from the Bloomington Stock Center (stock no. 4968) and has not been characterized at the molecular level 33 . aub N11 has a 154-bp deletion, resulting in a frameshift which is predicted to add 16 novel amino acids after residue 740 (refs 34, 35). Dicer-2 L811fsX is a loss-of-function allele induced by EMS that has a sequence variant at residue 811 resulting in a stop codon 36 . It is homozygous, viable and fertile. Quantification of TSE. When TSE is incomplete, variegation is observed because 'on/off' lacZ expression is seen between egg chambers: that is, egg chambers can show strong expression (dark blue) or no expression, but intermediate expression levels are rarely found. TSE was quantified as previously described 5 by determining the percentage of egg chambers with no expression in the germ line. Deep sequencing analyses. Small RNAs from hand-dissected ovaries were cloned using the DGE-Small RNA Sample Prep Kit and the Small RNA Sample Prep v.1.5 Conversion Kit from Illumina (libraries 1 to 5), following the manufacturer's instructions, or using the TruSeq (TM) SBS v.5 Kit at Fasteris (http://www.fasteris. com/) (libraries 6 to 8). Libraries 1 to 5 were sequenced using an Illumina Genome Analyzer II and libraries 6 to 8 were sequenced using an Illumina Hi-Seq 2000. Sequence reads in fastq format were trimmed from the adaptor sequence 59-TCGTATGCCGTCTTCTGCTTG-39 (libraries 1 to 5) or 59-CTGTAGG CACCATCAAT-39 (libraries 6 to 8) and matched to the D. melanogaster genome release 5.43 using Bowtie 37 , as well as to the sequences of the P-element constructs P{lArB} (FlyBase accession FBtp0000160) and P{lacW} (FlyBase accession FBtp0000204). Only 19-30-nucleotide reads matching the reference sequences with 0 or 1 mismatch were retained for subsequent analysis. For global annotation of the libraries (Supplementary Table 2), we used release 5.43 of fasta reference files available in FlyBase, including transposon sequences (dmel-all-transposon_r5.43.fasta) and release 18 of miRNA sequences from miRBase (http://www.mirbase.org).
Sequence length distributions, small RNA mapping and frequency maps were generated using in-house Python scripts and R (http://www.r-project.org/) to